Anthropic's $1.5 billion copyright settlement, fair use of AI issues remain unresolved

On July 20, 2026, a U.S. federal judge approved Anthropic to settle a $1.5 billion copyright lawsuit

filed by multiple authors accusing the AI startup of using pirated books to train its Claude model. The settlement, originally reached in 2025, is the largest copyright settlement known to involve generative AI and the clearest demonstration of the costs copyright risks can bring to large language model development companies.

The settlement does not answer the industry's biggest legal question: whether using copyrighted material to train AI models is fair use. Instead, reconciliation attempts to give a value to this uncertainty. For developers, investors and publishers, the case provides the first way to assess the real cost of the consequences of disputes related to training data.

The lawsuit was called "Bartz et al. v. Anthropic PBC"(Case No. 3:24-cv-05417) and was heard by the U.S. District Court for the Northern District of California. Although the case may not establish a binding legal precedent, it could have broad implications for issues surrounding copyright and the way such disputes are resolved.

Fair use issues remain unresolved

The core of conflicts of laws lies in the concept of "fair use." The U.S. Copyright Office defines it as a legal principle that allows the use of copyrighted material without the permission of the copyright owner under certain circumstances. The court will judge each case based on factors such as the purpose of use, the nature of the copyrighted work, the quantity used, and the impact on the market for the original work.

Whether this protection applies to AI training remains uncertain. Peter Henderson, a JD/PhD student at Stanford University and co-author of "Basic Models and Reasonable Use", said the issue has not yet been resolved. "There is uncertainty about how such litigation will develop."-- Peter

This uncertainty is not theoretical. Researchers led by Henderson found that with simple modification of the prompt words or a small number of derivatives, GPT-4 was able to provide long sections of text from Dr. Seuss's "Oh, Where You're Going" and "Harry Potter and the Philosopher's Stone", even with some minimal modifications to the wording of the prompt words.

This finding and other similar results were used to demonstrate that certain AI models retain protected expressions rather than just learning patterns from training material. The U.S. Copyright Office reached a similar conclusion, pointing out that whether AI training can be classified as fair use depends on the specific facts of each case rather than by general legal principles.

Benchmark for future AI litigation

The legal process began in 2023, when some authors claimed Anthropic used illegal books to train its system. In 2025, the company decided to settle for $1.5 billion after Judge William Alsup concluded that AI training could indeed be considered fair use, but other claims related to obtaining pirated books could still be heard. On July 20, 2026, Judge Araceli Martinez-Olguin finally approved the agreement, making it the largest copyright settlement in the field of generative AI.

This timing is important because U.S. courts are currently hearing a number of high-profile lawsuits, such as the New York Times lawsuit against OpenAI and Microsoft, multiple lawsuits filed by multiple writers against Meta, and lawsuits filed by different publishers, writers and media organizations. In addition to these cases, Anthropic's settlement provides an immediate benchmark, but legal issues remain unresolved.

Markets are waiting and see

As far as the AI industry is concerned, this settlement means that copyright risks are turning from intangible legal concerns to actual business costs. Companies and institutions have invested billions of dollars in chips, cloud systems and computing power, and soon access to defensible training data will become another competitive advantage.

All of this points to a bigger change: To succeed in AI, you need not only to focus on computing power and model performance, but also to have good legally protected data to deliver success. Perhaps the court will one day rule on the boundary between fair use and permission. At the same time, Anthropic is likely to remain an unofficial benchmark for the cost of industry uncertainty and reminds many AI companies that settlements are sometimes less risky than waiting for a court decision.